Design-Thinking-Driven Variational Autoencoder for Human Activity Classification

This study introduces a design-thinking-driven approach aimed at enhancing the accuracy and robustness of Human Activity Recognition (HAR) using sensor data from Internet of Things (IoT) devices. The proposed algorithm harnesses the synergy of a Variational Autoencoder-Based Dimensionality Reduction Technique with deep learning methods for human activity classification. HAR plays a pivotal role in categorizing activities, especially in domains like elder care and fitness tracking. This research work systematically assesses the model’s performance across diverse datasets, including WISDM, HAPT, HAR, and KU-HAR, utilizing performance evaluation metrics such as accuracy, recall, precision, and F1-Score. The results underscore the potential of this design-thinking-infused approach to significantly enhance the precision and reliability of activity prediction in IoT-based HAR applications. The study introduces a design-thinking-driven approach by combining Variational Autoencoder-Based Dimensionality Reduction with deep learning techniques for enhancing Human Activity Recognition (HAR).